How would you research and develop techniques for model alignment, value learning, and interpretability at Character.AI?